【作者】 王晴；

【导师】 尹苏皖；

【作者基本信息】 四川大学 ， 工程硕士（专业学位）， 2025， 硕士

【摘要】 根据国际航空运输协会(IATA)的预测,中国预计将在2035年成为全球最大的航空市场,航空需求的增加势必会进一步增大机场运行负荷。为了提升场面运行效率、减少航班延误,学术界与业界已开展深入的研究,提出并实施了多种创新的运营概念、流程以及决策支持工具,包括机场协同决策系统(A-CDM)、高级场面滑行引导与控制系统(A-SMGCS)、离场管理系统(DMAN)以及场面管理系统(SMAN)等。这些自动化系统中,滑行规划问题包括实时滑行时间预测和动态滑行路径优化已成为其中的核心组成部分之一。基于此,本文旨在基于Kriging模型设计并实现一套适用于复杂机场的场面滑行优化框架。该方法的优势在于通过采用数学结构,它能够在不直接使用复杂模型的情况下,进行高效的分析、优化或决策,进而实现对模型粒度、复杂性和准确性的平衡,能够为场面实时滑行优化的实现提供可靠的理论支撑。本文的主要研究内容包括:(1)基于Ordinary Kriging模型构建了适用于复杂机场构型的滑行时间实时预测模型。为提升建模效率,提出了两种自适应算法:(i)自适应采样与填充策略(Adaptive infilling algo-rithm),能够自主筛选训练样本,从而大幅减少所需的训练样本量;(ii)超参数自适应选择算法(Kriging-AHP),用于加速模型训练过程。这两种自适应算法显著提升了模型在保持高预测精度的前提下的计算速度,使得所提出的Adaptive Kriging模型能够在实时预测飞机滑行时间方面,在模型粒度、复杂度与精度之间实现灵活平衡。与现有方法相比,该模型展现了更强的适应性,并具备较好的泛化潜力。(2)基于(1)的滑行时间预测结果构建实时动态滑行路径优化方法,基于滚动时间窗口策略,以“最小化目标时间窗内总滑行时间”为目标。为了全面评估优化效果,本文对比了多种路径策略,包括先到先服务(FCFS)、滚动时间窗口(Rolling Horizon)和滑动时间窗口(Sliding Time Window)策略,并深入分析了每种策略的优势与劣势,可为机场运营者深入理解各策略在不同情境下的表现,从而选择最适合的路径优化策略。此外,实时滑行路径优化分配的算法实现为未来A-SMGCS(第三阶段:滑行路径规划)的实施提供可靠的理论支撑,推动复杂机场动态调度和智能优化系统的发展。(3)为验证所提出方法的有效性,本文在北京首都国际机场(PEK)的高保真仿真环境下对基于Kriging的滑行时间预测与路径优化方法进行了测试。实验结果表明,所提出的优化算法框架在多个关键绩效指标(KPIs)方面显著优于现有方法。具体地,基于30分钟滚动时间窗口优化,平均滑行时间减少5.3%,航班准点率提升14.4%,滑行冲突次数降低18.3%,跑道头队列时间减少48.7%。基于上述优化结果,所提出方法在运行成本效益、环境可持续性、空管人员工作负荷以及旅客服务满意度等方面展现出良好的应用前景与综合价值。更多还原

【Abstract】 The International Air Transport Association(IATA)forecasts that China will become the largest aviation market in the world by 2035.Such increased demand will place significant pressure on air-port surfaces.To improve operational efficiency and reduce flight delays,both academia and indus-try have conducted extensive research and proposed and implemented various innovative operational concepts,processes,and automated systems,including Airport Collaborative Decision-Making(A-CDM),Advanced Surface Movement Guidance and Control Systems(A-SMGCS),Departure Man-ager(DMAN),and Surface Manager(SMAN).The taxiway planning problem,including real-time taxi time prediction and dynamic taxi path optimization,is one of the core components among these automated systems.Therefore,this thesis proposes a framework for hub-airport surface taxiing plan-ning based on an Ordinary Kriging model considering distinct airport topologies.The proposed ap-proach facilitates efficient approximations and real-time decision-making in complex airport scenarios by leveraging mathematical structures,thereby eliminating the need for overly complex models.In doing so,it achieves a well-balanced trade-off between model granularity,complexity,and accuracy.The objectives of the work are as follows.(1)Develop a metamodeling framework,named Kriging,designed to predict taxiing time in real-time.To address computational challenges,two algorithms are proposed:(i)an adaptive sampling and infill strategy that significantly reduces the required training sample size,and(ii)an adaptive hyperparameter selection algorithm aimed at speeding up the training process.These two adaptive algorithms considerably enhance the model’s computational speed while maintaining high prediction accuracy.As a result,the proposed Adaptive Kriging model effectively achieves a flexible balance between granularity,complexity,and accuracy in real-time taxi time prediction.The results of experi-ment demonstrates that the proposed model exhibits superior adaptability and potential generalization,outperforming existing work.(2)Based on the taxi time prediction results obtained in(1),a real-time dynamic taxi route opti-mization method is developed.This is employed based on a rolling horizon approach with the objec-tive of minimizing the total taxi time within the target time window.To comprehensively evaluate the effectiveness of the proposed optimization,this work compares multiple routing strategies,including First-Come-First-Serve(FCFS),Rolling Horizon,and Sliding Time Window approaches.A thorough analysis of the strengths and weaknesses of each strategy is conducted to provide airport operators with deeper insights into their performance under various operational scenarios,thereby facilitating the selection of the most suitable routing strategy.Furthermore,the algorithmic implementation of real-time taxi route optimization offers reliable theoretical support for the realization of A-SMGCS Level 3(Route Planning),advancing the development of dynamic scheduling and intelligent opti-mization systems at complex airports.(3)To validate the effectiveness of the proposed method,the Kriging-based taxi time prediction and path optimization methods were tested in a high-fidelity simulation environment at Beijing Capi-tal International Airport(PEK).The experimental results demonstrate that the proposed optimization algorithm framework significantly outperforms existing methods across several key performance in-dicators(KPIs),achieving the following improvements:a 5.3%reduction in average taxi time,an18.3%decrease in taxi conflicts,a 48.7%reduction in runway queue time,and a 14.4%increase in flight punctuality.Based on these optimization results,the proposed method demonstrates promising application prospects and comprehensive value in terms of operational cost efficiency,environmental sustainability,air traffic controller workload,and passenger service satisfaction.更多还原